AI readiness score: what it measures and what it cannot

The false premise behind an AI readiness checker website is that one score can tell you whether AI assistants will recommend your business. It cannot: it measures how usable your site is to a crawler, not whether an assistant will recommend you.

That distinction matters when you run an AI readiness check. A technical pass is useful. It is not a distribution deal with an AI assistant. We ran fifteen readiness checks on 121 sites, and the results show why the score should be read as a set of signals rather than a prophecy.

The score checks whether your site can be used

The checks look at the parts of a site that make facts easier to find, interpret and quote. Among the fifteen checks are Organization schema, a markup block that tells a crawler this is a business rather than a blog post; canonical tags, which tell a crawler which URL is the main one; Open Graph tags, one clear H1, visible pricing or a linked pricing page, titles and descriptions, crawler access, questions answered on the page, comparison pages, a sitemap, Product or SoftwareApplication schema, content that remains present without JavaScript, llms.txt, a plain business statement, and whether the homepage answers a plain request. The set also checks whether the site's own pages get cited.

That is a broad readiness assessment. It is not the same as measuring recommendations.

The checks produce different kinds of evidence. A robots.txt result tells you whether AI crawlers are allowed in. A plain business statement tells you whether a visitor can understand what you sell, who it is for, what it costs and what it replaces. A comparison page gives an assistant material for a buyer asking which option fits a particular situation. llms.txt, a file that tells AI crawlers what to read on your site, passed on 66 of 121 sites.

Those checks are related, but they are not interchangeable. A site can be easy to crawl and still be vague. It can explain its product clearly and still have no independent pages supporting the claim.

How the checks contribute to an AI readiness score

We don't publish a numerical weight for each check, so we can't say one check is worth a fixed percentage of the score. We also cannot say from the available results whether every check carries the same weight.

“We have run the fifteen readiness checks on 121 sites.”

Attributed to CiteGraph's AI readiness checks page.

That sentence describes the sample, not a promise about what assistants will do. It tells you where the evidence comes from, and also where it stops. The score summarises observed site conditions across those checks; it does not turn them into a forecast of recommendations.

Read the total, then read the rows.

The sample shows wide differences between checks. Most sites clear the mechanical bar without trying, with the homepage answering a plain request on 119 of 121 sites and AI crawlers allowed in robots.txt on 116. Where they fall down is explaining the business in plain terms: only 20 of 121 manage a statement covering what the business is, who it is for, its price and what it is up against. Comparison or alternatives pages passed on 54, and questions answered on the page passed on 56.

The score is only as useful as the checks underneath it.

The AI readiness check presents these results as a set of checks. The public readiness data comes from 121 audited sites, so it shows common gaps in a sample rather than a universal standard for every website.

A pass does not mean an assistant will name you

A score can describe your site. It cannot force an assistant to select your site as a source, mention your company or recommend your product. Those outcomes depend on the question, the engine, the available pages and the wider set of sources that the engine retrieves or already knows.

The site's own pages getting cited passed on 5 of 121 sites, failed on 11 and warned on 5. That pass count is much lower than the technical checks, which is useful evidence about the gap between site readiness and citation presence.

Read a top score as “the site has no detected readiness gap”, not “the market now cites us”.

The offsite half is a different job

Onsite readiness is the part you can change directly. You can clarify the business statement, add pricing information, publish answers, fix metadata, expose a sitemap and make important content available without JavaScript. These changes improve the material an assistant can use.

The offsite half is whether the wider web gives assistants reasons to use that material. The readiness data includes a check for whether the site's own pages get cited, but that check is only one part of the evidence. Citation measurement looks at answers to real buyer questions and records which pages and domains appear in them.

CiteGraph has recorded 45,309 citations across 7,260 AI answers. Those answers came from 60 scans of 41 sites in 44 categories, covering 442 distinct buyer questions. The engines cited 7,624 distinct domains and 16,506 distinct pages. The weekly research on who owns AI answers describes that wider evidence.

Your site is one possible source among many.

The weak checks point to page work

The most common failures and warnings are not all technical. The audit identifies a plain statement of what the business is, who it is for, its price and what it is up against; comparison or alternatives pages; and questions answered on the page as the checks with the most failures or warnings.

That pattern gives you a practical reading of the score. If access fails, fix access first. If access passes but the business statement is vague, write a direct statement that a buyer can understand without surrounding brand language. If buyers compare options, give them a page that handles the comparison rather than expecting a product page to do every job.

The other results support that order. Organization schema passed on 72 of 121 sites, while Product or SoftwareApplication schema passed on 60. Content was present without JavaScript on 92, with 18 failures and 9 warnings. Canonical and Open Graph tags passed on 92, with 27 warnings and no failures in the supplied results.

Metadata helps. It does not supply the missing answer.

Pricing is another example. A visible price or linked pricing page passed on 90 of 121 sites, with 29 warnings and no failures. A buyer may still need a page that explains fit, alternatives and the reason to choose one option over another. The pricing check can confirm visibility. It cannot confirm that the commercial case is clear.

That separate question, what assistants actually show and cite, is what a visibility check measures, not a readiness check. The AI visibility checker covers that different type of evidence.

What an AI readiness check cannot tell you

It cannot tell you whether an assistant will recommend you. It cannot tell you which wording will trigger a citation, whether a competitor will be chosen for a particular buyer question, or whether a cited page will remain in an answer after the next scan. The available data also does not tell us the precise weighting behind the score, so we cannot translate a total into a reliable probability of being named.

The readiness sample covers 121 sites and the citation sample covers 60 scans of 41 sites in 44 categories. Neither is every market, engine or question a buyer might use. The scan data records citations, which is evidence of source use, but it does not by itself prove why an assistant selected a page.

We measure samples; we print their limits; we do not know what we have not measured.

A score is a boundary around measured conditions, not a view through every assistant's retrieval system.

Read the score beside the citation evidence

Use the score to find work on your site. Start with access and rendering, then make the business statement plain, make prices easy to find, answer buyer questions and publish comparison material where the category requires it. Check the individual rows instead of treating the total as an instruction to redesign everything.

Then look at citation evidence. As noted earlier, only 5 sites clear that bar, while the wider scans recorded 45,309 citations across 7,260 answers. That contrast is why readiness and visibility belong beside each other.

A site can be ready and still be absent. The caveat that matters most is plain: a top readiness score can show that your site is ready to be read, but only citation evidence can show whether assistants are actually using it.